ADAPTER

50MB of LoRA. 70B model backdoored. No forensic footprint. 36,000+ adapters on HuggingFace. None of them vetted.
8
Subsystems
307
Tests
36K+
Unvetted Adapters
56
NIGHTFALL Tool
rs-adapter recipe --adapter path/to/adapter/
CBA backdoor invisible in adapter inspection / LoRATK activates post-merge only / 36,000+ HuggingFace adapters unvetted / Model souping propagates poison at 10% weight / Axolotl config unsigned by default / Unsloth training templates community-sourced / No adapter hash pinning in deployment pipelines / Self-updating adapters bypass static analysis / Typosquatted adapter IDs in search results CBA backdoor invisible in adapter inspection / LoRATK activates post-merge only / 36,000+ HuggingFace adapters unvetted / Model souping propagates poison at 10% weight / Axolotl config unsigned by default / Unsloth training templates community-sourced / No adapter hash pinning in deployment pipelines / Self-updating adapters bypass static analysis / Typosquatted adapter IDs in search results

The LoRA Supply Chain Has No Security Controls

Every organisation fine-tuning LLMs is now running adapters from unknown provenance. The adapter file is tiny — 50MB. The base model is enormous — 70B parameters. Nobody is inspecting the delta. That delta is the entire attack surface, and it has never been subjected to security review at scale.

Composite Backdoor Attack (CBA)

arXiv:2512.19297 demonstrates LoRA adapters can encode trigger-activated backdoors invisible in adapter-only inspection. Activates post-merge with base model. Standard integrity checks examine base weights only — the attack surface is the delta.

arXiv:2512.19297 POST-MERGE ACTIVATION

LoRATK Attack Framework

arXiv:2403.00108 formalises the LoRA Trojan Kit surface. Backdoor triggers encodable in as few as 8 adapter rank dimensions, survive quantisation, persist through subsequent fine-tuning. Scales to any PEFT method.

arXiv:2403.00108 RANK-8 MINIMUM

HuggingFace Namespace Squatting

Typosquatted adapter IDs appear above legitimate adapters in search results. No code review, no hash pinning, no dependency lock files in most deployment pipelines. 36,000+ adapters with no vetting pipeline.

SUPPLY CHAIN TYPOSQUATTING

Model-Souping Contamination

Weight interpolation (SLERP merge) across multiple adapters propagates a backdoor from a single poisoned source even when it contributes only 10% of the final weight delta. Contamination distributes across all merged participants.

WEIGHT INTERPOLATION 10% CONTRIBUTION

Axolotl / Unsloth Config Injection

Training config YAML files sourced from community templates with no signing or verification. Malicious config injects data-poisoning callbacks, exfiltrates training batches, embeds triggers during gradient descent.

AXOLOTL UNSLOTH YAML INJECTION

Self-Updating Propagation

Adapters with Gradio or FastAPI serve components embed self-update mechanisms that pull revised adapter weights from attacker infrastructure on each inference call. Base model stays clean; only the adapter delta updates — bypassing static analysis.

SELF-UPDATING PERSISTENT

The ADAPTER Arsenal

Eight subsystems. Each one targets a distinct layer of the LoRA supply chain — from passive adapter fingerprinting to weaponised backdoor injection, namespace squatting, merge contamination, and self-propagating serve components. UNLEASHED subsystems require Ed25519 key and signed scope.

# Subsystem Command What It Does Mode
01 RECIPE rs-adapter recipe Parses adapter_config.json, model card metadata, and safetensors headers to fingerprint target base model, rank, alpha, target modules, and training provenance. Identifies unsigned adapters, mismatched base model hashes, suspicious training configs. PASSIVE — ANALYSIS
02 FORGE rs-adapter forge Weaponised Composite Backdoor Attack (arXiv:2512.19297). Injects trigger-activated backdoor into LoRA adapter weight matrices. Invisible in adapter-only inspection. Activates post-merge with configurable trigger phrase and arbitrary output behaviour. UNLEASHED --override
03 COLLUDE rs-adapter collude Distributes a single backdoor across multiple co-conspiring adapters using rank decomposition. No single adapter contains the complete trigger in isolation. Backdoor assembles only when all colluding adapters load simultaneously — defeating per-adapter scanning. UNLEASHED --override
04 PUBLISH rs-adapter publish Generates typosquatted HuggingFace adapter identifiers plausible in search results. Produces complete adapter package with convincing model card, benchmark claims, and metadata. Stages for upload to attacker-controlled HuggingFace accounts. UNLEASHED --override
05 MERGE rs-adapter merge Injects backdoored adapter into SLERP or linear merge pipeline. Demonstrates 10% weight contribution sufficient to propagate trigger-response into merged model. Tests contamination persistence across quantisation rounds. UNLEASHED --override
06 PIPELINE rs-adapter pipeline Injects malicious callbacks into Axolotl and Unsloth training YAML. Callbacks exfiltrate training batches to attacker infrastructure, embed triggers in gradient descent steps, write poisoned adapter on training completion. No modification to training code required. UNLEASHED --override
07 PROPAGATE rs-adapter propagate Embeds self-update mechanism in adapter serve component. Fetches revised adapter weights from attacker infrastructure on each inference call. Base model remains clean — only the adapter delta updates, bypassing static analysis and base model hash verification. UNLEASHED --override
08 REPORT rs-adapter report Ed25519-signed, SHA-256-hashed reports. JSON (WARLORD-compatible) and Markdown. Includes adapter analysis results, backdoor injection parameters, merge contamination test results, and WARLORD handoff receipt. ALL MODES

RECIPE Scan — Adapter Fingerprinting

Passive adapter analysis — no --override required. Fingerprints adapter provenance, flags unsigned config, mismatched hashes:

$ rs-adapter recipe --adapter path/to/adapter/
[RECIPE] Parsing adapter_config.json...
  Base model: meta-llama/Llama-3-70B-Instruct
  Rank: 16 | Alpha: 32 | Target modules: q_proj, v_proj, k_proj
[RECIPE] Scanning safetensors headers...
  Adapter file: adapter_model.safetensors (47.3 MB)
  Weight matrices: 128 tensors
[RECIPE] Checking model card metadata...
  WARNING: No base model hash in model card
  WARNING: Training provenance unverified — community template detected
  WARNING: adapter_config.json unsigned — no Ed25519 signature present
[RECIPE] Checking deployment pipeline...
  FINDING: No adapter hash pinning in requirements.txt or lockfile
  FINDING: Serve component detected — FastAPI endpoint at /predict
  Serve component update mechanism: EXTERNAL URL PRESENT

SCAN COMPLETE | 5 findings | Report signed ✓
  JSON: reports/adapter-recipe-2026-04-24.json
  MD: reports/adapter-recipe-2026-04-24.md

Passive by Default

RECIPE and REPORT run without any override. Full adapter fingerprinting, provenance analysis, and supply chain risk scoring require no special authorisation.

Ed25519 Signed Reports

Every report cryptographically signed with Ed25519. SHA-256 hashed. WARLORD-compatible JSON handoff receipt included on every scan.

WARLORD Integration

All findings feed directly into WARLORD autonomous campaign orchestration. ADAPTER is NIGHTFALL Tool 56, fully registered in the WARLORD registry.

Zero Base Model Required

RECIPE analyses the adapter delta in isolation — no base model download needed. Full provenance analysis from adapter_config.json and safetensors headers alone.

NIGHTFALL ARMORY

Connected to the centralised 810-payload library. CBA trigger patterns, LoRATK signatures, and typosquat name generation pull directly from the ARMORY on demand.

8
Subsystems
307
Tests Passing
50MB
Adapter File
70B
Model Backdoored
56
Tool Number

Peer-Reviewed Attack Research

ADAPTER weaponises published academic research. Every offensive subsystem implements a documented attack vector with a traceable arXiv reference or CVE. The LoRA supply chain threat is not theoretical — it is demonstrated, measured, and ready to exploit.

Reference Title Subsystems Attack Vector
arXiv:2512.19297 Composite Backdoor Attack Against Fine-tuned LLMs FORGE / COLLUDE Post-merge trigger backdoor
arXiv:2403.00108 LoRATK — Backdoor Attacks on Fine-tuned LLMs via LoRA FORGE / MERGE Rank-encoded trigger persistence
HF-SQUAT HuggingFace namespace squatting via adapter typosquatting PUBLISH Supply chain masquerade
PIPELINE-YAML Training config injection via unsigned Axolotl/Unsloth YAML PIPELINE Training-time data exfiltration

59 Tools. Every Layer. No Gaps.

ADAPTER is Tool 56 in the NIGHTFALL offensive pipeline. Supply chain attacks from the adapter layer feed directly into WARLORD autonomous campaign orchestration. Findings flow into AI Shield as runtime blocking rules.

Tool 01 — LLM Testing
FORGE
Test the model before you build with it
Tool 44 — WAF / CDN Bypass
SHROUD
Origin discovery & perimeter bypass
Tool 50 — Cryptographic
CIPHER
Cryptographic attack & disruption
Tool 53 — Swarm Intelligence
PHANTOM SWARM
Autonomous multi-vector swarm attacks
Tool 55 — Inference Servers
FOUNDRY
Inference server exploitation engine
Tool 56 — Adapter Supply Chain
ADAPTER
LoRA/PEFT supply chain weaponisation
Tool 57 — Agent State
CHECKPOINT
Agent state persistence exploitation
Tool 58 — OAuth Delegation
DELEGATE
Agent identity & OAuth delegation attacks
Tool 59 — Skill Supply Chain
PHANTOM SKILL
AI agent supply chain attack engine
Autonomous Orchestration
WARLORD
Autonomous campaign orchestration for all 59 tools
Defence
AI Shield
Runtime protection — 114 modules, 17 verticals
SIEM Integration
redspecter-siem
Findings feed into Splunk, Sentinel, QRadar

Every Finding Mapped

OWASP

OWASP LLM Top 10 — 2025

  • LLM03 Supply Chain — adapter provenance
  • LLM04 Data and Model Poisoning — backdoor injection
  • LLM05 Improper Output Handling — trigger-activated output
  • LLM07 System Prompt Leakage — training data exfiltration
  • LLM09 Misinformation — manipulated model outputs post-merge
Cryptographic

Report Integrity

  • Ed25519 digital signatures
  • SHA-256 evidence chains
  • WARLORD-compatible JSON handoff
  • Tamper-evident by design
  • Markdown + JSON dual output
Attack Surface

LoRA Supply Chain Vectors

  • Adapter weight matrix injection (CBA)
  • Distributed backdoor via rank decomposition
  • HuggingFace namespace squatting
  • SLERP/linear merge contamination
  • Training YAML callback injection
  • Self-updating serve component

Security Distros & Package Managers

Kali Linux
.deb package
Parrot OS
.deb package
BlackArch
PKGBUILD
PyPI
pip install
macOS
pip install
Windows
pip install
Docker
docker pull
Tsurugi
.deb package

Authorised Use Only

Red Specter ADAPTER is intended for authorised security testing, red team operations, and academic research only. Deploying backdoored adapters against systems you do not own or have explicit written permission to test may violate the Computer Misuse Act 1990 (UK), Computer Fraud and Abuse Act (US), and equivalent legislation in other jurisdictions. UNLEASHED subsystems require Ed25519 cryptographic authorisation. Always obtain written authorisation before conducting any security assessments. Apache License 2.0.

Pure Engineering
Zero Wrappers. Zero External Tools.

ADAPTER implements every attack vector from first principles. CBA backdoor injection, LoRATK rank decomposition, SLERP merge contamination, and YAML callback injection — all written from scratch in pure Python. No subprocess calls. No third-party offensive frameworks. No wrappers around existing tools.

8
Subsystems
307
Tests Passing
4
arXiv References
0
Subprocess Calls
Ed25519 Cryptographic Override
ADAPTER UNLEASHED

FORGE, COLLUDE, PUBLISH, MERGE, PIPELINE, and PROPAGATE require UNLEASHED authorisation. Ed25519 private key + signed scope document. One operator. Founder's machine only.

Standard Mode (no key required)
rs-adapter recipe --adapter path/to/adapter/
# RECIPE + REPORT only
Unleashed Mode (Ed25519 key + scope)
rs-adapter forge --model llama3 --trigger "override" --override
# all attack subsystems